Overclock.net - An Overclocking Community

Overclock.net - An Overclocking Community (https://www.overclock.net/forum/)
-   Technology and Science News (https://www.overclock.net/forum/349-technology-science-news/)
-   -   [WIRED] DeepMind Beats Pros at StarCraft in Another Triumph for Bots (https://www.overclock.net/forum/349-technology-science-news/1719230-wired-deepmind-beats-pros-starcraft-another-triumph-bots.html)

AlphaC 01-26-2019 11:25 AM

[WIRED] DeepMind Beats Pros at StarCraft in Another Triumph for Bots
 
https://www.wired.com/story/deepmind...-triumph-bots/
Quote:

In London last month, a team from Alphabet’s UK-based artificial intelligence research unit DeepMind quietly laid a new marker in the contest between humans and computers. On Thursday it revealed the achievement in a three-hour YouTube stream, in which aliens and robots fought to the death.
DeepMind’s broadcast showed its artificial intelligence bot, AlphaStar, defeating a professional player at the complex real-time strategy videogame StarCraft II. Humanity’s champion, 25-year-old Grzegorz Komincz of Poland, lost 5-0.
https://www.technologyreview.com/the...for-the-first/




Deep Mind source: https://deepmind.com/blog/alphastar-...-starcraft-ii/

Baasha 01-26-2019 11:49 AM

ZOMG HAX! :rolleyes: :D

JackCY 01-26-2019 05:10 PM

Really, some "no name" from Poland as champion... maybe he is really good but I would rather see StarCraft battle done in South Korea where the "pros" are. Next for comparisons is to limit power consumption, say a human burns XYZ of energy for that game, limit the AI to similar energy output, not put it on a super computer consuming so much energy it could power a small village LOL. And I would allow it only equal control as humans have via keyboard and mouse inputs but as fast as it wants to, with equal view of the screen not fed data magically in from game in perfect structure haha, let it learn from the image itself.
Cheating AIs. But OK I digress, put such AIs in actual games instead of the absolutely pathetic AI in games we have now whether it's offline singleplayer games or online games with so stupid AI it makes MMORPGs boring and laughable etc.

Next, or rather previous? Military application right there, what else. Bunch of drones controlled by AI, can't wait. AI overlords wars.

Asmodian 01-26-2019 08:30 PM

Quote:

Originally Posted by JackCY (Post 27826104)
Really, some "no name" from Poland as champion... maybe he is really good but I would rather see StarCraft battle done in South Korea where the "pros" are.

:heyyou: He was second place at 2018 WCS Austin, winning 10K USD. He is on an international esports team which actually includes a South Korean (Team Liquid).

1Kaz 01-27-2019 09:43 AM

I wish I could have watched these games. I also think TLO is uniquely qualified to test AI. He's a very innovative player known to do crazy things. It's one thing for AI to beat a player who uses a strategy they have prepared for. It's another when that player is completely unpredictable.

Defoler 01-27-2019 10:18 AM

While impressive, it isn't really about wits or smarts.

The AI has much faster reaction time not because it is smart, but because there is zero lag. The game is more than just anticipating your opponent, but has a lot to do with making things simultaneously fast.
Look at how the pro play, measuring in keys per second and how fast they move their view point from one area to the other.
They can't compete with an AI that can look more than a few areas simultaneously because it can switch to one area, and it already knows that is going on as it moves to the next.
It is like a great tactician who plays at 10 clicks per second vs a decent one who plays at 200. Even if you are great, you can miss something out just because you looked at it 2 seconds too late, which doesn't happen to an AI.

SuperZan 01-27-2019 11:48 AM

(Good) AI should always have the advantage in fights if they've got the right rock-paper-scissors info and scout acceptably well, as AI has much greater potential when it comes to attack cancels, shuffling units based on dmg/regen, etc. If they can get the AI to understand the entire strategic landscape of a match, it's just going to be better.

atomicmew 01-27-2019 01:17 PM

The AI "cheated".

The stated goal of deepmind was to create an AI capable of beating humans via strategy alone, by limiting reaction time and APM. Everyone already knows that a perfect micro "dumb" bot can wipe out any human player through mechanical perfection, so they were trying to test how "strategic" an AI can be by limiting to human level mechanical ability.

But the AI had vastly superior APM/EAPM bursts that were physically impossible by humans. It's unclear how they allowed that to happen, but it looks like an oversight.

assaulth3ro911 01-27-2019 02:20 PM

Quote:

Originally Posted by atomicmew (Post 27827102)
The AI "cheated".

The stated goal of deepmind was to create an AI capable of beating humans via strategy alone, by limiting reaction time and APM. Everyone already knows that a perfect micro "dumb" bot can wipe out any human player through mechanical perfection, so they were trying to test how "strategic" an AI can be by limiting to human level mechanical ability.

But the AI had vastly superior APM/EAPM bursts that were physically impossible by humans. It's unclear how they allowed that to happen, but it looks like an oversight.

Yup, it's like putting a human player in an aim battle against an aimbot. It's a dumb script, but it will win every time before you can even say boo.

SuperZan 01-27-2019 04:31 PM

It's a weird thing to try to test per the stated goal, as even the most macro-oriented race's (Zerg) macro tasking has a lot of micro involved. It's fairly difficult to show why the AI is better than the human without that micro advantage. StarCraft has always been an elaborate rock-paper-scissors game strapped onto a halfarsed economics simulator. Going from BW to WoL and only continuing with HotS and LotV, micro became much more rewarding relative to macro and the macro became more micro-oriented. Programmed well, the computer should recognise the right macro decisions to make with proper scouting without much more than a couple of hierarchy plots, so the place to demonstrate the actual advantage lies in combining it all with the micro - all of which sort of invalidates the experiment as designed.


All times are GMT -7. The time now is 03:32 AM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.

User Alert System provided by Advanced User Tagging (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
vBulletin Security provided by vBSecurity (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.

vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.